Regional Science and Urban Economics 29 (1999) 519–539

Climate consumption and climate pricing from 1940 to 1990 Michael I. Cragg, Matthew E. Kahn* Columbia University, Department of Economics and SIPA, New York, NY 10027, USA Accepted 3 December 1998

Abstract This paper studies trends in US climate consumption and climate compensating differentials. Three findings emerge. First, the population has shifted so that the consumption of warmer winter climate has risen for both working families and more dramatically for senior citizens. Second, there has been a rise in rental capitalization. Third, earnings capitalization has declined. Between 1960 and 1990, the relative price of climate has increased for senior citizens and fallen for working families.  1999 Elsevier Science B.V. All rights reserved. Keywords: Regional migration; Climate; Compensating differentials JEL classification: R23

1. Introduction Climate is the leading example of a spatially tied local public good. In the United States, regional migration to the South and West has changed the average person’s climate bundle. This paper documents trends in climate consumption from 1940 to 1990. It quantifies how much households are paying for climate through compensating differentials capitalized into housing prices and earnings. By estimating repeated hedonic regressions in 1960–1990, we construct a Laspeyres climate price index and compare how climate expenditures have changed over time for working and non-working families. *Corresponding author. 0166-0462 / 99 / $ – see front matter  1999 Elsevier Science B.V. All rights reserved. PII: S0166-0462( 98 )00046-5

520

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Understanding the changes in the pricing and consumption patterns for climate is important for a variety of reasons. First, credible climate valuation estimates would inform policy makers on the benefits of mitigating Global Warming (Nordhaus, 1996).1 Second, warmer winter climate is an input in health capital production. Thus, estimates of trends in climate prices are useful in determining an overall consumer price index that reflects changes in the price of achieving a given quality of life. The Boskin Report has addressed incorporating quality of life trends into the US government’s official CPI.2 By estimating the same hedonic specification at four points in time, in 1960– 1990, we contribute to the hedonic literature by studying the robustness and the plausibility of compensating differential estimates (Graves et al., 1988). Empirical studies have found that climate is capitalized into the hedonic rental and wage gradient (Gyourko and Tracy, 1991; Blomquist et al., 1988).3 The price of climate, as measured by tied earnings and rents, is partially determined by supply and demand in the regional housing and labor markets.4 We study how regional income convergence has affected estimated climate prices (Barro and Sala-i-Martin, 1991, 1995; Oi, 1997; Topel, 1986).5 The paper proceeds by first measuring how the population’s climate consumption has changed over the last 50 years. We estimate how the price of climate has evolved between 1960 and 1990. The hedonic price estimates are used to construct climate expenditure indices for working families and retired households. We find that climate is capitalized into earnings and rents in all four decades. The relative price of climate has changed over time for working and non-working families.

1 Using county level data from the 1980s, Nordhaus (1996) estimates wage regressions and concludes that plausible increases in carbon dioxide emissions could cause amenity losses of about 0.35% of aggregate US wages which represents about 0.17% of US GDP or $12 billion per year measured in 1995 prices and incomes. 2 Boskin et al. (1998) write ‘‘changes in the physical, social and economic environment may impose higher expenditures necessary to keep up with our previously achieved utility levels’’ (p. 20). 3 Based on 1980 census data, Gyourko and Tracy (1991) report statistically insignificant impacts of precipitation, cooling degree days, humidity, sunshine and wind speed on labor market and housing market compensation. Heating degree days is statistically significant indicating that in colder climates families are compensating with a combination of lower housing prices and higher earnings. They estimate that the annualized full price (i.e. annual earnings net of annual housing price) of a 1% rise in heating degree days is $22.58. 4 A variety of authors have examined the importance of factors which might affect regional labor demand and supply. For example, the work of Oi (1997) points to the role of the spread of air conditioning and Crandall (1993) documents trends in the spatial location of manufacturing. To examine the regional effects of shifts in defense spending, exchange rate shocks and oil price shocks see Davis et al. (1995). Costa (1998) presents a novel analysis of trends in leisure consumption. 5 Between 1960 and 1990, the correlation of average rents and earnings in a state is 0.61 and 0.70, respectively.

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

521

Climate expenditure has increased more for the retired than for one worker or two worker families because there has been a shift in climate capitalization from earnings to rents. In the conclusion, we discuss the implications of our findings for an aging population whose willingness to pay for climate is likely to grow over time.

2. Increased climate consumption from 1940–1990 In the United States, households are consuming more temperate climate over time. Due to the migration to the South and West from the North, average February temperature consumption has increased from 34.68 in 1960 to 36.98 in 1990.6 Migration to the South and West has accelerated in the second half of the 20th Century. In both 1890 and 1940, Darke County, Ohio, was the nation’s median location such that half the population lived east of this county and half lived west of this county, and half lived north of this county and half lived south of it. Between 1940 and 1990, the median location shifted to Monroe, Indiana, which lies approximately 110 miles west and 75 miles south of Darke, Ohio (Statistical Abstract of the United States, 1995, in US Bureau of the Census, 1995). To explore changes in climate consumption between 1940 and 1990, we tabulated each state’s average February temperature and weighted the tabulation by the state’s population in 1940 and 1990.7 This yields a measure of the fraction of people who are consuming at most the corresponding average February temperature in 1940 and 1990. This cumulative distribution function for February temperature in 1940 and 1990 is presented in Fig. 1. To interpret this graph, it is relevant to note that as the cumulative distribution function shifts right, people are experiencing warmer climates. As the cumulative distribution function becomes more bowed, the consumption of temperature becomes more equal. Fig. 1 shows that in 1940, half of the population lived in states where the average February temperature was less than 308. By 1990, fewer than 40% of people experienced average February temperatures below freezing. Fig. 1 shows that the nation is consuming more temperate climate. It also demonstrates that the consumption of climate is becoming less equal. There has been no change in temperature consumption in the bottom fifth of the population while the top 80% are consuming much more temperate climate.

6 We calculate the national weighted average using state average temperature weighted by the state’s percentage of the national population. 7 Using Public Use Micro Census data (Appendix A) we identified each person’s state of residence and merged on average February temperature for that state. Sorting the data by February temperature yields the cumulative distribution function which is plotted in Figs. 1 and 2.

522

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Fig. 1. Climate consumption CDF in 1940 and 1990.

Fig. 2 presents the same climate distributions but limits the sample to people over age 64. Relative to the whole population, this group has more sharply increased its temperature consumption (Rogers, 1992; Muth, 1991). To document that the population has shifted to warmer areas, we examine migration patterns. As discussed in Appendix A, Census of Population and Housing microdata identifies both a household’s current state of residence and also their state of residence five years earlier. Rather than present a 48 3 48 state transition matrix, we assign states to one of four groups based on February temperature and present a 4 3 4 transition matrix which links climate rankings of one’s origin and destination state.8 Table 1 shows that in 1940, 39% of movers younger than age 65 and 47% of movers age 65 and up increased their consumption of February temperature. By 1990, movers are much more likely to increase their climate consumption. Almost 60% of the older movers and almost 50% of working age movers increased their consumption of February temperature while only a quarter of the movers relocated in a colder location. Migration research has found a consistent relationship between net migration 8 Each of the 48 states are assigned to one February temperature category. The states with mean temperature below 23.8 are assigned to the first quartile. States with February temperature between 23.8 and 33.1 are assigned to the second quartile and states with temperature between 33.1 and 39.6 are assigned to the third quartile. States with temperature above 39.6 are assigned to the fourth quartile. This partition is such that each of the groups includes 12 states.

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

523

Fig. 2. Climate consumption CDF for age$65 in 1940 and 1990.

and warm winter climate (Graves, 1979, 1980).9 Mueser and Graves (1995) present county level evidence based on data from 1950–1990 and find that controlling for county per capita income there is a positive correlation of net migration and February temperature but they find that this correlation is growing smaller over time. Cragg and Kahn (1997) use 1990 Census data to estimate a state level discrete choice model of cross-state migrants’ locational choice. Controlling for economic opportunity, we find evidence that warmer climates are a normal good and that willingness to pay for warmer winter climate rises with age.10

3. Measuring the change in climate prices between 1960 and 1990 In this section, we estimate national hedonic earnings and rental regressions to study how climate prices have evolved from 1960 to 1990. Our climate compensating differentials estimates contribute to national studies of inter-location 9 Berger and Blomquist (1992) present a more comprehensive study linking migration to a broader range of quality of life indicators. They study whether migrants are overcoming migration costs to move to the counties where their quality of life metric ranked as high. 10 Analogous to Quigley (1982) methodology, they exploit the non-linearity of the migrant’s budget constraint to identify the parameters in the utility function.

524

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Table 1 Climate 52year transition matrices for migrants State quartile rank in 1940

State quartile climate rank in 1935

All age groups

1st

2nd

3rd

4th

1st

0.24

0.24

0.16

0.06

2nd 3rd 4th 0.39

0.31 0.23 0.22 Fraction in upper triangle 1st

0.29 0.24 0.22 0.27 2nd

0.30 0.20 0.35 Fraction on diagonal 3rd

0.12 0.21 0.60 0.33 4th

1st 2nd 3rd 4th 0.39

0.25 0.31 0.23 0.21 Fraction in upper triangle 1st

0.25 0.30 0.24 0.21 0.28 2nd

0.16 0.30 0.19 0.35 Fraction on diagonal 3rd

0.06 0.12 0.21 0.60 0.34 4th

Fraction in lower triangle

1st 2nd 3rd 4th 0.47

0.17 0.22 0.24 0.38 Fraction in upper triangle

0.15 0.22 0.24 0.39 0.22

0.12 0.23 0.24 0.41 Fraction on diagonal

0.05 0.15 0.19 0.61 0.31

State quartile rank in 1990

State quartile climate rank in 1985 1st

2nd

3rd

4th

1st 2nd 3rd 4th 0.49

0.13 0.23 0.21 0.42 Fraction in upper triangle 1st

0.16 0.19 0.23 0.42 0.26 2nd

0.10 0.24 0.20 0.46 Fraction on diagonal 3rd

0.09 0.18 0.26 0.47 0.25 4th

1st 2nd 3rd 4th 0.48

0.15 0.25 0.22 0.38 Fraction in upper triangle 1st

0.16 0.21 0.24 0.39 0.26 2nd

0.11 0.24 0.21 0.44 Fraction on diagonal 3rd

0.09 0.18 0.25 0.47 0.26 4th

1st 2nd 3rd 4th 0.57

0.07 0.13 0.15 0.64 Fraction in upper triangle

0.12 0.11 0.17 0.60 0.22

0.07 0.17 0.17 0.59 Fraction on diagonal

0.09 0.17 0.28 0.46 0.20

Fraction in lower triangle Age,65

Fraction in lower triangle Age.65

All age groups

Fraction in lower triangle Age,65

Fraction in lower triangle Age.65

Fraction in lower triangle

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

525

quality of life (Roback, 1982; Rosen, 1979; Gyourko and Tracy, 1991; Blomquist et al., 1988; Gabriel et al., 1996). The starting point for all hedonic studies is an equilibrium assumption that spatial variation wages, rents, and amenities have adjusted such that the marginal person and firm are indifferent across locations. Assuming that migration costs are low, spatial price variation represents an equilibrium compensating differential for non-market spatially tied goods such as climate. As factors affecting the supply and demand for labor or housing change over time, hedonic prices might shift. We test four hypotheses concerning climate price dynamics: • Hypothesis 1 : Climate compensating differentials have not changed over time. • Hypothesis 2 : Rental prices have risen and earnings have fallen in warm winter locations. • Hypothesis 3 : The effect of July temperature and humidity have fallen over time as cheaper air conditioning mitigates their impact on wages and rents. • Hypothesis 4 : There has been a shift of climate capitalization from earnings to rentals such that families now pay for climate through rental capitalization. In 1920, 39.8% of the US population lived in the Southern and Western regions. By, 1994 this percentage has grown to 56.6%. This population shift might bid up rents and lead to lower earnings in more temperate places. Hypothesis 2 studies this conjecture. Climate is a key input in the household production of comfort. In the past, Southern states were a bundle of pleasant winters and uncomfortable summers. The air conditioner has allowed households to consume warm winter without suffering from a location’s warm and humid summer months. This innovation affects both consumption of quality of life and the marginal productivity of firms located in such areas. Roback (1982) discusses the effects of a productivity enhancing amenity. Over time, earnings could rise in the South due to increased local labor demand (Oi, 1997). Our hedonic estimates provide new insights into how the interaction of new products (air conditioning) and spatially tied local public goods affect compensating differentials. The fourth hypothesis focuses on how the rise in women’s labor force participation and the growth of the number of retired households migrating to the Sun Belt between 1960 and 1990 has affected how much different types of households pay for local public goods. Two worker households have an incentive to choose areas rich in local amenities which are capitalized into rents not earnings. The retired have the opposite incentive. Our analysis builds on work by Graves and Waldman (1991). They noted that the retired should be attracted to local amenities that are mainly capitalized into wages rather than rents. Since the retired do not work, they face a lower price for such local public goods relative to single earner households and especially relative to dual earner households.

526

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Migration may affect the relative capitalization of local public goods into earnings and rents. To study these hypotheses, we quantify how hedonic climate prices have evolved by estimating climate compensating differentials at four points in time. We estimate national reduced form cross-sectional earnings and home price regressions as in Eqs. (1) and (2).11 log(earnings ijt ) 5 Bt Xit 1 gt Zj 1 eijt

(1)

log(rent ijt ) 5 ftVit 1 ct Zj 1 Uijt

(2)

where i is a person or housing unit, j indicates the geographical unit of analysis (either a state or a metropolitan area), and t is time and X and V represent characteristics of the worker and the housing unit and Z are characteristics of the geographical location that the worker or home are tied to. In estimating Eqs. (1) and (2), we assume that there is a national market in housing and labor such that the price per unit of human capital varies over time but not across regions. The cross-market quality of life literature has assumed that controlling for observed characteristics all spatial variation in earnings and rents is generated by differences in quantities of amenities (Zs) not differences in prices across space (see Gyourko and Tracy, 1991; Blomquist et al., 1988). Several issues arise in estimating Eqs. (1) and (2). First, to be able to go back in time as far back as 1960 we use the state rather than the metropolitan area as the geographical unit of analysis. We present hedonic estimates based on micro data with state level local public goods merged to these records using data from 1960, 1970, 1980 and 1990. In addition, we present hedonic rental regression estimates using 1980 and 1990 Census microdata where the geographical unit of analysis is the metropolitan area. In estimating Eqs. (1) and (2), we represent climate along four dimensions that include a state’s historical average February temperature, July temperature, percentage days of sunshine and humidity (see Appendix A).12 Estimates of g and c are used to generate climate price indices based on implicit capitalization into wages and rents. Climate is an attractive amenity to quantify since its cross-state

11 Previous hedonic studies have estimated hedonic wage rather than earnings regressions. Wages are constructed by dividing earnings by hours worked. Thus, the typical hedonic study’s dependent variable is constructed as the difference in log of earnings and log of hours. Wage estimates will be biased towards 0 if climate and leisure are complements. If base salaries are constant across space, but salaried workers reduce their labor supply in nice places, then hourly wages will be higher in nice areas. By using earnings rather than wages as our dependent variable, our method calculates the total private consumption sacrificed in consuming a location’s climate. 12 We only use the four climate proxies, sunshine, humidity, July and February average temperature because other measures like heating and cooling days are very highly correlated with these.

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

527

variation is larger than its within state variation.13 As an additional control for local public goods, we include the geographical area’s mean education level. Rauch (1993) has documented, using 1980 census data, that controlling for a rich set of worker attributes, that individuals living in cities where the average person is highly educated the individual worker earns higher wages and pays higher rents. Following Rauch, we construct for each geographic area in each year of the data the area’s mean education and include it as a local public goods amenity. To control for unobserved within state factors, we follow the recommendations of Moulton (1986); Gyourko and Tracy (1991) by computing Huber-White corrected standard errors where grouping is done by geographical area. Thus, the regression equations are estimated using generalized least squares.

3.1. Hedonic climate price estimates We estimate the same hedonic specification in 1960, 1970, 1980 and 1990 to generate climate price indices to test the hypotheses discussed above. Table 2 reports four separate hedonic estimates of the earnings regression in Eq. (1). In each decade the sample consists of all men ages 25–55, who worked between 1800 and 2600 hours and were in the manufacturing, services, government or finance, insurance and real estate industries. The dependent variable is the log of annual earnings net of state taxes measured in 1989$. The individual level controls for age, race, education, occupation, industry and marital status all have intuitive coefficient signs. Since they have been well documented in the labor literature, we focus on the market capitalization of the local public goods included in the specification. At the bottom of each column, we report an F-statistic indicating that at the 5% significance level that we can reject the hypothesis that the four climate variables all are statistically insignificant from 0 in all four regressions. The results in Table 2 indicate that climate compensating differentials capitalized into earnings have changed over time. In 1960, February temperature was statistically significant in lowering earnings. All else equal, an extra standard deviation of February temperature (10.48) lowers earnings by 3.1%. This capitalization remained roughly steady in 1970 and 1980. By 1990, however, February temperature is no longer capitalized into earnings. This indicates that workers in warm winter locations were not paying through lower earnings in 1990 as much as they were in 1960. The compensation for July temperature has fallen over time. In 1960, an area with an extra standard deviation of July temperature paid 3.6% higher earnings. This estimate is just barely statistically significant. In each passing decade the coefficient on July temperature shrunk and by 1990 it is actually 13

Ideally, we would want to exploit within state variation in climate by merging climate data by an individual’s city location. Unfortunately to create a consistent repeated cross-sectional panel, the ‘‘smallest’’ geographical unit we can adopt is the state. Cross-state hedonic estimates of crime’s capitalization or school quality would be much more suspect.

528

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Table 2 Hedonic earnings regressions 1960–1990

Professional occupation dummy Service industry dummy Public administration dummy Finance, insurance, and real estate (FIRE) dummy College graduate Years of education Age (Age)2 White dummy Married dummy State average education level Average February temperature Average July temperature Sunshine Humidity Constant Observations R2 F-test for joint significance of the four climate variables

1960

1970

1980

1990

0.078 (12.063) 20.237 (216.931) 20.103 (28.309) 20.047 (4.367) 0.046 (25.480) 0.072 (7.571) 0.058 (31.034) 20.001 (224.452) 0.284 (15.298) 0.134 (25.568) 0.139 (6.807) 20.003 (22.533) 0.006 (1.681) 0.000 (0.367) 0.004 (2.344) 5.803 (12.910) 91 712 0.30 2.39*

0.077 (9.878) 20.171 (220.319) 20.045 (.3.641) 0.010 (0.786) 0.114 (10.634) 0.050 (33.723) 0.061 (28.908) 20.001 (224.955) 0.253 (13.504) 0.120 (26.539) 0.156 (7.369) 20.002 (21.425) 0.005 (1.397) 20.003 (21.846) 0.004 (2.278) 5.771 (11.931) 107 186 0.27 4.40**

0.144 (33.447) 20.263 (219.086) 20.108 (26.616) 20.065 (25.791) 0.019 (3.213) 0.054 (33.337) 0.073 (26.887) 20.001 (223.544) 0.190 (15.672) 0.097 (15.840) 0.145 (5.549) 20.003 (21.914) 0.004 (1.367) 20.001 (20.654) 0.004 (2.446) 5.331 (9.575) 109 973 0.20 2.79**

0.170 (27.984) 20.231 (216.900) 20.098 (24.723) 0.012 (0.565) 0.068 (7.847) 0.061 (34.914) 0.060 (30.198) 20.001 (223.560) 0.146 (10.153) 0.093 (22.608) 0.186 (4.289) 20.000 (20.088) 20.004 (20.861) 0.003 (1.243) 0.005 (2.550) 5.023 (5.736) 101 138 0.26 2.78**

T-statistics in parentheses. The dependent variable is the log of annual earnings net of state taxes (units 1989$). The regressions include males aged 25–55, who worked between 1800 and 2600 h and were in the manufacturing, services, government or FIRE industries. The omitted category is a non2white, non2married worker who is not a college graduate, not a professional occupation and works in the manufacturing industry. The regressions are estimated using GLS. ** p2value,0.01. * p2value$0.01 and p-value,0.05.

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

529

negative but statistically insignificant. Unlike February and July temperature, sunshine’s price trends do not tell a consistent story over time. Only in 1970 was its coefficient statistically significant. In 1970, an extra standard deviation of sunshine lowered earnings by 2.4%. In 1960, 1970, 1980 and 1990, earnings are statistically significantly higher in more humid areas. Humidity’s coefficient is 0.004 in 1960, 1970 and 1980 and rises only slightly to 0.005 in 1990. This means that an extra standard deviation of humidity required earnings compensation of 4.1%. While July temperature’s compensating differential has fallen over time, humidity’s compensation has not. Across all four decades we find evidence in support of Rauch (1993) finding that local educational attainment offers positive externalities which boost earnings. We find a larger effect of local education than Rauch. A state’s average education level has a statistically significant effect on a man’s earnings. In 1990, an extra standard deviation of state level educational attainment (0.37 years) raises a person’s earnings by 6.9%. One explanation for this is that our sample includes men who do and do not live in metropolitan areas. Areas with higher levels of average education are likely to be more urbanized. Thus, the average state education level reflects both the urbanization effect and the positive human capital externality effect. The rental regressions measuring the degree of climate capitalization into housing rental prices are presented in Table 3. The regressions include controls for the number of rooms and year built dummies to control for the vintage of the housing stock. The housing structure controls have intuitive coefficient estimates and are highly statistically significant. The F-tests for each of the four regressions presented in the columns of Table 2 indicate that in 1960, 1970 and 1990, we can reject the hypothesis that climate is not capitalized into rentals at the 1% level. Surprisingly, we fail to reject this hypothesis in 1980. For the individual climate variables, we find that in 1960 and 1970 areas with warmer February temperatures featured lower rental prices. In 1980, February temperature was not capitalized into rental prices. In 1990, rents were higher in warmer winter states. In 1960, an extra standard deviation of February temperature lowered rents by 10% while in 1990 an extra standard deviation raised rents by 7.3%. Trends in July temperature capitalization are the mirror opposite of the February temperature price trends. In 1960, rents were higher in high July temperature states while in 1990 rents were lower. In 1990, an extra standard deviation of July temperature lowered rents by 13%. Sunshine’s price coefficient was statistically insignificant from 0 in 1960–1980. In 1990, an extra standard deviation of sunshine raised rents by 10.5%. Humidity consistently raises rents.14 This counter-intuitive coefficient estimate indicates that an extra standard deviation of humidity raises rents by 6.2%. Similar to Rauch (1993) findings, rents are

14

Gyourko and Tracy (1991) also report a positive impact of humidity on rents.

530

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Table 3 Hedonic rental regressions 1960-21990

Rooms Unit is 3-25 years old Unit is 6-210 years old Unit is 11-220 years old Unit is 21-230 years old Unit is 31-240 years old

1960

1970

1980

1990

0.059 (6.593) 20.084 (26.609) 20.218 (215.184) 20.389 (221.499) 20.463 (224.627) 20.591 (221.045)

0.066 (7.768) 20.015 (21.073) 20.132 (24.732) 20.366 (211.323) 20.546 (215.353) 20.689 (220.100)

0.077 (7.845) 0.027 (1.552) 0.014 (0.480) 20.028 (20.684) 20.196 (23.698) 20.299 (4.980) 20.354 (26.931)

0.377 (9.457) 20.010 (22.882) 0.013 (1.595) 20.002 (20.578) 0.007 (2.202) 3.339 (3.341) 170 267 0.23 4.12*

0.403 (9.416) 20.008 (22.118) 0.007 (0.964) 20.003 (20.676) 0.006 (2.032) 3.305 (3.329) 332 412 0.33 4.40**

0.354 (8.083) 20.001 (20.498) 20.005 (20.806) 0.003 (0.992) 0.004 (1.529) 3.573 (3.622) 238 616 0.20 1.35

0.074 (8.128) 0.018 (0.792) 20.190 (4.882) 20.216 (29.058) 20.181 (27.237) 20.238 (27.957) 20.293 (27.501) 20.311 (28.661) 0.528 (6.420) 0.007 (1.625) 20.025 (23.314) 0.013 (2.766) 0.010 (2.782) 1.390 (0.860) 263 964 0.19 16.6**

Unit is 41-250 years old Unit is 501 years old State average education level Average February temperature Average July temperature Sunshine Humidity Constant Observations R2 F 2test for joint significance of the four climate variables

The dependent variable is the log of real contract rent measured in 1989$. The omitted category is a 1 year old rental unit. T2statistics are presented in parentheses. The regressions are estimated using GLS to control for within-geographic area correlation of the error terms. ** p2value,0.01.

higher in states with higher levels of average education. In 1990, an extra standard deviation of state average educational attainment raises rents by 19.5%. Two concerns with the hedonic rental regressions presented in Table 3 are the statistically insignificant 1980 results and the limited number of housing unit control variables included in the specification. To further study recent trends in climate prices, we use data from 1980 and 1990 1% PUMS. Restricting the sample to renters who live in metropolitan areas, we run hedonic rental regressions based on Eq. (2). We are able to control for a richer set of housing controls including

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

531

how many bedrooms the unit has and whether the unit is a single detached structure, single attached, or part of an apartment building complex. In order to conserve on space, we suppress the coefficients on unit type in the results presented in Table 4.15 Table 4 reports the hedonic estimates based on the metropolitan area data sets. Two findings emerge. Unlike the 1980 results presented in Table 3, the 1980 results based on the metropolitan area households indicates that climate is capitalized into rents. The F-test for the four climate proxies is highly statistically significant. In 1980, July temperature lowers rents and sunshine raises rents. The 1990 coefficient estimates reported in Tables 3 and 4 look quite similar. It is also relevant to note that the metropolitan area human capital variable has a smaller positive coefficient in Table 4 than the education results reported in Table 3.16 The findings presented in Tables 2–4 clearly reject the hypothesis that climate Table 4 Hedonic rental regressions 1980-21990 1980

Rooms Bedrooms Unit is 3-25 years old Unit is 62-10 years old Unit is 112-20 years old Unit is 212-30 years old Unit is 312-40 years old Unit is 412-50 years old Unit is 501 years old Metropolitan area median education level Average February temperature Average July temperature Sunshine Humidity Constant Observations R2 F 2test for joint significance of the four climate variables

1990

b

T2stat

b

T2stat

0.099 0.005 0.015 20.019 20.044 20.166 20.254 20.306

16.421 0.382 1.171 21.017 21.985 26.221 27.630 210.226

0.064 0.002 20.023 0.007 20.002 8.423 197 048 0.19 11.05**

1.787 1.186 25.744 3.832 21.104 13.384

0.025 0.045 0.295 0.163 0.148 0.176 0.126 0.087 0.029 0.131 0.001 20.033 0.014 0.005 7.294 257 937 0.12 13.45**

3.043 4.237 10.956 4.912 4.502 5.250 3.814 2.467 0.766 1.746 0.332 26.499 4.884 2.329 6.287

The sample includes all rental units in metropolitan areas. The dependent variable in each regression is the log of real annual contract rents. Eight fixed effects which control for the number of units in the structure are also included in the regressions but are suppressed. The regressions are estimated using GLS to control for within-geographic area correlation of the error terms. ** p2value,0.01. 15

The full set of coefficient estimates are available on request. We have also collected other metropolitan area level local public goods such as proxies for crime. Similar to Gyourko and Tracy (1991), we find that including crime in the housing hedonic regressions yields a counter-intuitive positive coefficient. These results are available on request. 16

532

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

compensating differentials have not changed over time.17 We find evidence in support of our second hypothesis. All else equal, warm February areas feature rising rental prices and rising earnings. Surprisingly, we find no evidence in favor of the hypothesis that humidity’s capitalization has diminished over time.18 In the next section, we present our methodology for creating climate expenditure indices over time.

4. Climate expenditure indices The hedonic regressions yield implicit prices for climate. We combine these prices with climate quantity consumed to construct climate expenditure indices by decade. We also construct a Laspeyres price index which is useful for determining how the price of a fixed climate ‘bundle’ has changed over time. In addition, we study how the total price of climate for workers and non-workers has changed over time. Non-workers pay for climate solely through the housing market while workers may pay through the housing market and through earnings compensating differentials. Our first step in constructing climate expenditure indices is to present climate consumption (based on the four climate variables included in the hedonic regressions). In 1960–1990, we calculate how much February temperature, the average person under the age of 65 consumes. The results reported in Table 5 are normalized with Minnesota’s climate as the numeraire. Thus, if everyone lived in Minnesota mean climate consumption would be 0. Table 5 shows that the average person under age 65 consumed 18.83 extra February temperature than Minnesota’s. Between 1960 and 1990, average February consumption increased by 2.38 for people under age 65 and by 2.88 for people over age 65. Both groups have increased their consumption of July temperature, sunshine and slightly increased their consumption of humidity. The hedonic regression estimates of Eqs. (1) and (2) reported in Tables 2–4 yield a time series of climate prices. We combine information on the quantity of climate consumed and the prices paid to construct annual climate expenditure. Conceptually, the earnings expenditure on climate in each state is the sum across all workers of the price capitalized in earnings for each type of amenity multiplied by the quantity of the amenity in the state. National expenditure is simply the sum across the 48 states. Climate expenditure is calculated based on Eq. (3). 17

We know of no hedonic studies that have explored the robustness and plausibility of hedonic prices over the interval we have explored. 18 Since July temperature is highly correlated with February temperature it may be difficult to isolate the individual climate prices for the two amenities separately.

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

533

Table 5 Average climate consumption in 1960–1990 by age group Average climate consumption t 5

O s*(climate 2 climate jt

j

MN

)

Age,65

1960

1970

1980

1990

Average February temperature Average July temperature Sunshine Humidity Age$65

18.830 3.291 7.629 22.738 1960

19.398 3.362 7.965 22.758 1970

20.473 3.653 8.531 22.825 1980

21.083 3.778 8.935 22.939 1990

Average February temperature Average July temperature Sunshine Humidity

18.453 3.251 7.308 22.616

19.417 3.475 7.770 22.694

20.721 3.823 8.322 22.666

21.216 3.857 8.592 22.824

This table presents the population weighted average climate consumption for each demographic group in each year. The weights (s) represent a state’s share of the population. Minnesota’s climate level is normalized to 0.

O s ( p9 z 48

expenditure 5

1j

1j 1j

2 p1j z MNj )

(3)

j 51

Total climate expenditure pj 1 5z j 1 in each state j is calculated by using the hedonic price estimates pj 1 , which are simple transformations of the hedonic coefficients from our log-linear hedonic regressions.19 Climate expenditure in any year is the weighted sum across all states using each state’s population share, s j 1 , as weights. The expenditure index represents how much the average person pays for climate relative to consuming Minnesota’s climate. Table 6 reports climate expenditure (based on Eq. (3)) in 1960–1990 for workers and for renters. The first two rows of Table 6 report our expenditure estimates for men under age 65. In 1960, the average man was implicitly paid $450.6 a year for consuming the average climate (reported in Table 5) rather than Minnesota’s.20 Renter climate expenditure has increased over time. As shown in the right column of Table 6, in 1990, the average renter was spending $253 a year for consuming the mean climate relative to if he lived in Minnesota. Thus, between 1960 and 1990, annual climate payment through rents had increased by $703. While rental payments for climate have increased, worker compensating 19 By linearizing the hedonic climate price coefficients, we have for 1960–1990, four sets of earner and renter prices. In addition, there are the climate price estimates for 1980 and 1990 based on the metropolitan area regressions presented in Table 4. To construct the 1980 rental climate price, we take the average of the price based on the 1980 rental hedonic presented in Table 3 and the 1980 rental hedonic presented in Table 4. The 1990 rental hedonic prices is constructed in the same way. The general expenditure trends we report in Table 6 are robust with respect to the two indices we use. 20 Expenditures would be 0 each year if everyone chose to live in Minnesota.

534

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Table 6 Climate expenditure indices from 1960-21990

Age,65 Rental expenditure Earner compensation Age$65 Rental expenditure

1960

1970

1980

1990

2450.642 21144.52

2349.322 21577.33

27.653 21520.07

253.005 2179.757

2438.49

2345.864

15.946

233.452

Climate price index using the under age 65 population’s 1960 average climate consumption as the bundle Laspeyres index for renter non-worker 2450.642 2338.933 23.372 214.484 Laspeyres index for a one worker household 21144.52 21526.57 21415.34 2195.833 Annual climate expenditure measured in 1989$. Minnesota’s climate expenditure is normalized to $0 in each year. This table’s entries are based on climate price estimates which are created using the output in Tables 2–4 and the climate consumption quantities presented in Table 5. The Laspeyres index calculates how much the average 1960 climate bundle would cost as climate prices change across the decades.

differentials for climate have decreased. In 1960, the average worker implicitly paid $1144 to consume average climate rather live in Minnesota. By 1990, this payment had fallen to $179. This indicates that between 1960 and 1990, workers are paying less for avoiding Minnesota’s climate while renters are paying more. The next row of Table 6 reports climate expenditures for people over age 65. These expenditures are very similar to rental expenditures for people under age 65. Both groups pay the same rental prices for climate. In addition, at any point in time, people over age 65 and under age 65 are consuming roughly the same average climate bundle (see Table 5). The results indicate that in 1990 senior citizens were spending $670 more per year on climate than they did in 1960. Expenditure on climate can rise either because climate prices are rising or because people are consuming better climate. To decompose these changes in total expenditure, the final two rows of Table 6 report a climate Laspeyres index. Holding the spatial distribution of the population at its 1960 levels, we calculate what it would cost to ‘‘buy the old bundle at the new prices’’. The old bundle is the 1960 climate consumption. The index results show that changes in expenditure on climate are mostly due to prices changing not the quantities changing. The full payment for climate has actually fallen for working households over time but it has increased for those not in the labor force. The average senior citizen is paying $700 more for climate in 1990 relative to 1960. This is a significant amount of money which the current CPI does not reflect because it does not include the price of non-market local public goods such as climate. More households are choosing to consume a temperate climate and for nonworking households the price of consuming this bundle is rising. This suggests

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

535

that climate demand has increased.21 However, a rise in climate demand can not explain why rentals and earnings are growing in Southern areas (Topel, 1986). For firms to remain indifferent about staying in the South as rents rise, wage rates would have to fall to offset rising rental rates. Alternatively, in the face of rising rents, Southern firms would only be willing to stay if Southern labor productivity increased. Rising Southern labor demand has raised earnings in temperate climates. A number of authors have documented rising earnings and employment in the South.22 Oi’s thesis is that the adoption of the air conditioner has allowed increased worker productivity in the South.23 Thus, Southern economic development and not just rising climate demand is partially responsible for rising rents and earnings in this region.

5. Conclusion Between 1940 and 1990, the United States population has sharply increased its consumption of February temperature. This paper has studied trends in climate consumption. Through estimating hedonic regressions, we presented new climate price estimates between 1960 and 1990. Since climate is an important non-market good, it is important to quantify how its price has changed over time. Our analysis has yielded three important facts about how climate pricing and consumption have changed over the last 30 years. First, the population has shifted so that the consumption of warmer winter climate has risen for both working

21

There is a large literature exploring whether hedonic techniques can be used to infer structural demand parameters (Rosen, 1974; Bartik and Smith, 1987; Palmquist, 1991). Our hedonic estimates from 1960 to 1990 yield four price estimates and four quantity estimates. If supply side shocks to the housing market and labor demand shifts could be viewed as exogenous to climate demand then such shocks would provide information on how sensitive climate consumption is to changes in its price. Unfortunately, our four data points are clearly not enough information to estimate demand parameters using the hedonic ‘two step’ approach. 22 Mallick (1993) showed that in 1951 such Southern states as Mississippi and Arkansas had the lowest per-capita income in nation, but between 1951 and 1989 they ranked amongst the highest in growth. Nonetheless, in 1989 their level of per capita income is still amongst the lowest in the country. Oi (1997) presents additional evidence: from 1950 to 1990, the ratio of Southern per capita income relative to the national mean grew from 0.76 to 0.90. At the same time, there have been dramatic changes in employment and productivity. Between 1954 and 1987, the South’s share of manufacturing employment grew from 20.3% to 31.6% and value added per worker rose from $88.9 to $96.3. Additional research on regional convergence is presented in Amos (1991); Carlino and Mills (1996a), (1996b); Barro and Sala-i-Martin (1991), (1995); Sahling and Smith (1983); Treyz (1991); Mieszkowski (1979); Terkla and Doeringer (1991). 23 Oi points out that from 1970 to 1990, air conditioning rates in the South rose from 58 to 91% while at the national level they only rose from 44 to 70%. His discussion suggests that lower Southern wages and increasing productivity drove Southern economic development.

536

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

families and more dramatically for the elderly. Second, there has been a rise in rental capitalization. Third, earnings capitalization has declined so that the relative price of climate has increased for senior citizens and fallen for working families. Building on this paper’s findings, future work might explore the regional implications for the upcoming Baby Boomer retirement. A macro literature has explored how this large cohort affects national savings, the stock market, and aggregate real estate prices. We know of no work projecting how regional real estate prices will be affected by the large number of seniors pursuing climate (Welch, 1979; Berger, 1985; Mankiw and Weil, 1989). It is reasonable to predict that there will be increased capitalization of climate into rents and away from earnings.

Acknowledgements Early versions of this paper were presented at the AEA Winter Meetings in New Orleans in 1997 and at the November 1996 Regional Science meetings in Washington, DC. We thank the editor and two anonymous reviewers for helpful comments.

Appendix A The Public Use Micro Samples (PUMS) from the US Census of Population and Housing from 1940, 1960–1990, are the primary data sources used in our analysis. The PUMS is the only spatially representative sample which offers information on individual characteristics, earnings, rents, state of residence, and state of residence five years before. Two sets of PUMS data are used. The results in Tables 1–3 are based on PUMS data whose only geographical information is to indicate the resident’s state. In addition, the PUMS has answers to the question ‘Which state did you lived in five years earlier?’ for the 1940, 1980 and 1990 samples. We construct the share of migrants (disaggregated by age and education) who left each state five years earlier and the share who entered the state. The results in Table 4 are based on the 1% 1980 and 1990 PUMS which identifies for each person whether they live in a metropolitan area and which metropolitan area they live in. In the results in Table 4, we restrict the sample to only those individuals who live in a metropolitan area. Measures of annual earnings are available in the PUMS from 1960 to 1990. Our earnings regressions include males aged 25–55, who worked between 1800 and 2600 hours and were in the manufacturing, services, government or FIRE

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

537

industries. Our dependent variable is the log of annual earnings net of state taxes.24 The omitted category in the hedonic earnings regression is a man in a nonprofessional occupation employed in the manufacturing industry who is non-white and non-married. Measures of annual housing costs are available in the PUMS only from 1960 to 1990. The housing regressions only include renters and we use the log of annual contract rent as our dependent variable.25 All census summary statistics are available on request. The climate data sources are the Statistical Abstract of the United States, 1995 (US Bureau of the Census, 1995) and the US National Oceanic and Atmospheric Administration’s CD-ROM. The climate summary statistics are: Variable

Mean

Standard deviation

February temperature July temperature Sunshine Humidity 26

33.7 74.2 57.1 79.5

10.4 5.2 8.1 10.3

References Amos, O., 1991. Divergence of per capita real gross state product by sector 1963–1986, The Review of Regional Studies 21, 221–234. Barro, R., Sala-i-Martin, X., 1991. Convergence across states and regions, Brookings Papers on Economic Activity 1, 107–158. Barro, R., Sala-i-Martin, X., 1995. Economic Growth, McGraw Hill. Bartik, T., Smith, V.K., 1987. Urban amenities and public policy. In: Mills, Edwin S. (Ed.), Handbook of Regional and Urban Economics, Chap. 31, North Holland Press. Berger, M., 1985. The effect of cohort size on earnings growth an reexamination of the evidence, Journal of Political Economy 93(3) (June), 561–575. Berger, M., Blomquist, G., 1992. Mobility and destination in migration decisions, Journal of Housing Economics 2, 37–59. Blomquist, G., Berger, M., Hoehn, J., 1988. New estimates of quality of life in urban areas, American Economic Review 78, 89–107. Boskin, M., Dulberger, E., Gordon, R., Griliches, Z., Jorgenson, D., 1998. Consumer prices, the consumer price index and the cost of living, Journal of Economic Perspectives 12(1), 3–26.

24

While many states have non-linear income schedules, we found that the variation can be summarized using the lowest state tax bracket as a proxy. 25 By using only renters in the housing regressions, we avoid the issue faced by Gyourko and Tracy (1991); Blomquist et al. (1988) who include both renters and owners. They use monthly rent as the dependent variable and therefore must convert home values into implicit rents. 26 July afternoon average relative humidity (%). Average percentage of sunshine possible.

538

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

Carlino, G., Mills, L., 1996. Testing neoclassical convergence in regional incomes and earnings, Regional Science and Urban Economics 26, 565–590. Carlino, G., Mills, L., 1996. Convergence and the US states: A time-series analysis, Journal of Regional Science 36(4), 597–616. Costa, D., 1998. The rise of the leisured class. In: The Evolution of Retirement, Chap. 7, University of Chicago Press. Cragg, M., Kahn, M., 1997. New estimates on climate demand: Evidence from location choice, Journal of Urban Economics 42, 261–284. Crandall, R., 1993. Manufacturing on the Move, Brookings Institution, Washington, DC. Davis, S., Loungani, P., Mahidhara, R., 1995. Regional unemployment cycles, mimeo. Gabriel, S., Mattey, J., Wascher, W., 1996. Compensating differentials and evolution of quality of life among US states, Federal Reserve Bank of San Francisco, 96-07, June. Graves, P., Waldman, D., 1991. Multimarket amenity compensation and the behavior of the elderly, American Economic Review 81 (December), 1374–1381. Graves, P., Murdoch, J., Thayer, M., Waldman, D., 1988. The robustness of hedonic price estimation: Urban air quality, Land Economics 64 (August), 220–242. Graves, P. 1979. A life-cycle empirical analysis of migration and climate, by race, Journal of Urban Economics 6, 135–147. Graves, P. 1980. Migration and climate, Journal of Regional Science 20, 227–237. Gyourko, J., Tracy, J., 1991. The structure of local public finance and the quality of life, Journal of Political Economy 99 (August), 774–806. Mallick, R., 1993. Convergence of state per capita incomes: An examination of its sources, Growth and Change 24, 321–340. Mankiw, G., Weil, D., 1989. The baby boom, baby bust and the housing market, Regional Science and Urban Economics 19 (May), 235–258. Mieszkowski, P., 1979. Recent trends in urban and regional development. In: Mieszkowski, P., Straszheim, M. (Eds), Current Issues in Urban Economics, Johns Hopkins University Press, 1979. Moulton, B., 1986. Random group effects and the precision of regression estimates, Journal of Econometrics 32, 385–397. Mueser, P., Graves, P., 1995. Examining the role of economic opportunity and amenities in explaining population redistribution, Journal of Urban Economics 37, 176–200. Muth, R., 1991. Supply side regional economics, Journal of Urban Economics 29, 63–69. Nordhaus, W., 1996. Climate amenities and global warming, mimeo. Yale University, CT, USA. Oi, W., 1997. Welfare implications of invention. In: Bresnahan, Tim, Gordon, Robert (Eds.), The Economics of New Goods, NBER, University of Chicago Press. Palmquist, R., 1991. Hedonic methods. In: Braden, John B., Kolstad, Charles D. (Eds.), Measuring the Demand for Environmental Quality, Contributions to Economic Analysis, No. 198, Elsevier Science, New York. Quigley, J., 1982. Nonlinear budget constraints and consumer demand: An application to public programs for residential housing, Journal of Urban Economics, 177–201. Rauch, J., 1993. Productivity gains from geographic concentration of human capital: Evidence from the cities, Journal of Urban Economics 34(3) (November), 380–400. Roback, J., 1982. Wages, rents and the quality of life, Journal of Political Economy 90, 1257–1278. Rogers, A. In: Rogers, Andrei (Ed.), Elderly Migration and Population Redistribution in the US Elderly Migration and Population Redistribution, Wiley Press, 1992. Rosen, S., 1974. Hedonic prices and implicit markets: Product differentiation in pure competition, Journal of Political Economy 82 (January / February), 34–55. Rosen, S., 1979. Wage-based indexes of urban quality of life. In: Mieszkowski, Peter, Straszheim, Mahlon (Eds.), Current Issues in Urban Economics, Johns Hopkins University Press, Baltimore. Sahling, L., Smith, S., 1983. Regional wage differentials. Has the South risen again? Review of Economics and Statistics 65, 131–135.

M.I. Cragg, M.E. Kahn / Regional Science and Urban Economics 29 (1999) 519 – 539

539

Terkla, D., Doeringer, P., 1991. Explaining variations in employment growth: Structural and cyclical change among states and local areas, Journal of Urban Economics 29, 329–348. Topel, R., 1986. Local labor markets, Journal of Political Economy 94(3) pt. 2, S111–S144. Treyz, G., 1991. Causes of changes in wage variation among states, Journal of Urban Economics 29, 5–62. US Bureau of the Census, 1995. Statistical Abstract of the United States 1995, Washington, DC. Welch, F., 1979. Effects of cohort size on earnings: The baby boom babies’ financial bust, Journal of Political Economy 87(5) (October), S65–S98.